JP7490099B2 - 3D MULTI-SCALE MODELING SYSTEM AND METHOD - Patent application - Google Patents

Description

The present invention relates to a system and method for 3D multi-scale modeling.

Given the growing need to understand and control product and process behavior at multiple scales, multiscale modeling and simulation has emerged as a focal research area in applied science and engineering. Multiscale modeling refers to a modality of modeling in which multiple models at different scales are used to describe a system, e.g., a real-world system. In multiscale models, methods translate and transfer information at one scale to another scale, a process that may be termed "scale bridging" or "scale linkage". Scale bridging/linkage approaches exist and are of two general types: For materials/formulations to design engineering scale, existing approaches link data using empirical fitting; for design engineering to manufacturing scale, statistical process models exist.

Exemplary embodiments of computer-implemented methods and systems disclosed herein relate to three-dimensional (3D) multi-scale modeling and simulation of 3D systems. Although exemplary embodiments disclosed herein may be described with respect to complex composite systems, the embodiments are not limited thereto and may be used in a broader context for any material, mixture, or formulated and manufactured system.

Structures made of multiple materials (e.g., as a non-limiting example, materials having surface coatings) can be modeled via the exemplary embodiments disclosed herein to better represent real-world materials. Such exemplary embodiments can generate, as a non-limiting example, chemically and physically realistic models for mapping atomic material-scale attributes to measured attributes for a set of test coupons fabricated using processes from one or more materials. This is useful for improving simulations of material performance at the bulk scale and nanometer scale and beyond.

Such realistic models are also useful because they allow for studying non-local and non-equilibrium phenomena and understanding long-scale attributes such as, by way of non-limiting example, chemical decomposition in the environment. Simulations of such realistic models allow for, by way of non-limiting example, predictions of how materials (in computer-based virtual worlds) will change (e.g., aging, lifetime) and how they will perform in terms of design, safety, lifetime, etc. The results of such simulations can be used to modify the real-world systems on which the models are based. For example, embodiments can be used to predict failures of real-world systems and modify the design of such real-world systems.

According to an exemplary embodiment, a computer-implemented method generates a 3D multi-scale model of a 3D system. The computer-implemented method includes generating an artifact model that represents attributes, features, and artifacts of the 3D system at a given scale. The computer-implemented method further includes modifying a set of display models of the 3D system based on the generated artifact model. The modifying includes mapping the attributes, features, and artifacts to a display model of the set of display models at a higher or lower scale compared to the given scale. The mapping bridges (in time and space) a given display model of the set of display models at the given scale and a given display model of a representation model at the higher or lower scale. The computer-implemented method further includes automatically storing the artifact model in a database in association with the modified set of display models, thereby generating a 3D multi-scale model of the 3D system.

The computer-implemented method may further include automatically storing the model information in a database. The model information may represent starting information, training dataset information, learning method information, auxiliary data, measured or predicted data, or a combination thereof, to which the set of display models correspond. The model information may be associated in the database with the set of display models, artifact models, or a combination thereof. Generating the artifact models may include automatically identifying (determining) attributes, features, and artifacts via machine learning. The identification (determination) may be performed at a given scale. However, it should be understood that the identification (determination) may incorporate other scale information, such as from lower scale details or higher scale performance features. The lower scales are lower compared to the given scale, and the higher scales are higher compared to the given scale.

Machine learning may include, by way of non-limiting example, utilizing deep learning, adversarial learning, genetic or evolutionary methods, other modeling or segmentation classification approaches to modeling, or combinations thereof.

The computer-implemented method may further include performing machine learning on the systematic test results of the set of samples.

The computer-implemented method may further include controlling the machine learning according to at least one optimization criterion, such as, by way of non-limiting example, Pareto optimization determined in a closed loop or via a Pareto optimization analysis. The computer-implemented method may further include performing the machine learning iteratively based on a performance criterion (e.g., ultimate yield strength as a non-limiting example), a convergence threshold (e.g., 1%, 5%, etc., for non-limiting examples), a quality metric, a limit or limits, or a combination thereof. The computer-implemented method may further include determining whether the performance criterion, the convergence threshold, the quality metric, the limit or limits, or a combination thereof is satisfied via a closed loop or according to the at least one optimization criterion.

The computer-implemented method may further include (i) generating at least one of the set of display models based on a manufacturing process, and (ii) generating the set of display models by utilizing characteristics of a plurality of test coupons produced via the manufacturing process.

Each display model in the series of display models may be constructed at a different scale of a plurality of scales, the plurality of scales including the given scale. The computer-implemented method may further include generating a respective artifact model at each scale of the plurality of scales.

The computer-implemented method may further include, in a training phase, training a set of representation models based on at least one respective training data set. The computer-implemented method may further include, in a run phase, executing the 3D multi-scale model to produce predictions of onset of faults in the 3D system. The computer-implemented method may further include, in a validation phase, improving the accuracy of the predictions produced in the run phase by re-training the set of representation models and artifact models based on measurement data or prediction data input for the set of representation models.

The 3D system may be a building system, component, material, or structure. The structure may include i) a plurality of raw or intermediate materials, or ii) a mixture or formulation of a plurality of raw or intermediate materials.

The 3D system may be a real-world system. Each representation model in the series of representation models may be constructed at a different scale. The different scales may include a chemical-materials scale, a materials-materials scale, an engineering-design scale, an engineering-manufacturing-process scale, a system life scale, or a combination thereof.

According to another exemplary embodiment, a computer-based system for generating a three-dimensional (3D) multi-scale model of a 3D system comprises at least one memory and at least one processor coupled to the at least one memory. The at least one processor is configured to generate an artifact model indicative of attributes, features, and artifacts of the 3D system at a given scale. The at least one processor is further configured to modify a set of display models of the 3D system based on the generated artifact model. The set of modifications includes mapping the attributes, features, and artifacts to a display model of the set of display models at a higher or lower scale compared to the given scale. The mapping bridges a given display model of the set of display models at the given scale and a display model at the higher or lower scale. The at least one processor is further configured to automatically store the artifact model in a database in association with the modified set of display models, thereby generating a 3D multi-scale model of the 3D system.

Alternative computer-based system embodiments are similar to those described above in connection with the exemplary method embodiments.

It should be understood that the exemplary embodiments disclosed herein may be implemented in the form of a method, apparatus, system, or non-transitory computer-readable medium having program code embodied thereon.

The foregoing will become apparent from the following more particular description of exemplary embodiments, as illustrated in the accompanying drawings, in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed on illustrating the embodiments.

FIG. 1A is a block diagram of an exemplary embodiment of a computing environment in which a materials scientist uses a computer-based system to generate three-dimensional (3D) multi-scale models of 3D systems. FIG. 1B is a block diagram of an example embodiment of an artifact model and a set of display models. FIG. 2 is a flow diagram of an exemplary embodiment of a computer-implemented method for generating a 3D multi-scale model of a 3D system. FIG. 3 is a block diagram of an exemplary embodiment of a procedure for building a display model. FIG. 4 is a block diagram of an exemplary embodiment of features, process fabrication characteristics, and analysis elements that may be utilized by a complex procedure to build attribute models from nanoscale data having multiple material types. FIG. 5 is a block diagram of an exemplary embodiment of a procedure for building an attribute model from nanoscale data having one material type versus engineered part performance data. FIG. 6 is a block diagram of an example of the internal structure of a computer in which various embodiments of the present disclosure may be implemented.

An exemplary embodiment is described below.

Exemplary embodiments disclosed herein relate to three-dimensional (3D) multi-scale modeling and simulation of 3D systems and may be applied to areas using chemical, nanometer, and domain microstructures to predict and understand product performance and behavior, such as, by way of non-limiting examples, corrosion, degradation, creep, performance of surface protective coatings, matrix and fiber composites, complex mixtures and formulations, multi-stage flows, chemical reaction systems, or chemical process effects. It should be appreciated that while exemplary embodiments disclosed herein relate to 3D multi-scale modeling and simulation of complex composite systems, the embodiments are not limited to complex composite systems and may be applied, for example, to any building system, component, material, or structure.

At the core of a 3D multi-scale model are models and associated methods that couple processes at different scales, transform information at one scale, and communicate it to another. Such transformations and communication are referred to in the art as scale bridging (linkage). The scale linkage (bridging) frameworks (e.g., computer-based systems/architectures/methods) disclosed herein may be extendable to domains other than those disclosed above. Such frameworks may be deployed for domains based on the target material, test method, and/or time scale of observable features manufactured from the material. Approaches to scale linkage exist in the art and fall into two general categories:

(1) For materials, formulations, and design engineering scales, existing approaches use empirical fitting to link data. In contrast to (1), the exemplary embodiments disclosed herein may utilize machine learning models that account for fabrication and testing artifacts between properties determined at different scale levels.

(2) For design engineering to manufacturing scale, existing statistical process models cannot account for manufacturing and raw material variations and their impact on the process and final product quality. This is a significant approximation that the exemplary embodiments disclosed herein improve upon.

Both of the existing scale linkage approaches, i.e., (1) and (2) above, are methods for creating non-ideal models with little transferability or scalability, and therefore, such existing approaches create models that may be considered to be insufficient representations of real-world materials, their derived attributes, test samples, and/or products. In contrast to (1) and (2), exemplary embodiments of the systems and computer-implemented methods disclosed herein may use machine learning to generate chemically and physically realistic multi-scale models, for non-limiting examples, mapping atomic material scale attributes to attributes measured on a set of test coupons fabricated using a process from one or more materials. Such multi-scale models, as further disclosed below in connection with FIGS. 1A-B, are useful for improving simulations of material performance at the bulk scale, as well as simulations at nanometer scales and above.

Such multi-scale models are also useful for studying non-local and non-equilibrium phenomena and enabling understanding of long-scale attributes such as chemical decomposition in non-limiting example environments. Although the exemplary embodiments disclosed herein may be disclosed as applied to composite systems such as wind turbine blades as a non-limiting example, such exemplary embodiments are not so limited and may be used in a broader context for any material, mixture, or formulated and manufactured system. Structures including multiple materials may also be modeled via the exemplary embodiments disclosed herein to better represent real-world materials, such as those disclosed below in connection with FIG. 1A.

1A is a block diagram of an exemplary embodiment of a computing environment 100 in which a materials scientist 102 uses an exemplary embodiment of a computer-based system 104 to generate a three-dimensional (3D) multi-scale model 106 of a 3D system. A materials scientist is someone who studies the structure and chemical attributes of various materials for the development of new products or the enhancement of existing products. Of course, the computer-based system 104 is not limited to being used by materials scientists. The 3D system may be a building system, component, material, or structure, where the structure includes i) a plurality of raw or intermediate materials, or ii) a mixture or formulation of a plurality of raw or intermediate materials. The 3D system may be a real-world system, such as a wind turbine blade, as a non-limiting example.

In the exemplary embodiment of FIG. 1A, a materials scientist 102 uses a computer-based system 104 to generate a 3D multi-scale model 106 of a 3D system, i.e., a real-world (physical) wind turbine blade of a wind turbine, as a non-limiting example. The computer-based system 104 comprises at least one memory, such as a central processor unit 618 coupled to a memory 608, as disclosed further below with respect to FIG. 6, as a non-limiting example. The at least one processor coupled to the at least one memory. With further reference to FIG. 1A, the at least one processor of the computer-based system 104 is configured to generate an artifact model, such as an artifact model 110, as disclosed further below with respect to FIG. 1B, at a given scale.

FIG. 1B is a block diagram of an exemplary embodiment of an artifact model 110 of a 3D system and a set of display models 118 (i.e., display model _{0 ...} display model _N ), such as a wind turbine blade as a non-limiting example disclosed above and further below in connection with FIG. 1A . With reference to FIGS. 1A and 1B , at least one processor of the computer-based system 104 is configured to generate the artifact model 110 at a given scale. The artifact model 110 represents attributes 112, features 114, and artifacts 116 of the 3D system. The at least one processor is further configured to modify the set of display models 118 of the 3D system based on the generated artifact model 110. Modifying the set 118 includes mapping 120 the attributes 112, features 114, and artifacts 116 to display models in the set of display models 118 at a higher or lower scale compared to the given scale. The mapping 120 bridges a given display model of the set of display models 118 at a given scale with a display model of a higher or lower scale. The at least one processor is further configured to automatically store the artifact model 110 in association with the modified set of display models 118 in a database (not shown), thereby generating a 3D multi-scale model 106 of the 3D system, i.e., a wind turbine blade multi-scale model 108 of the wind turbine blade.

With further reference to FIG. 1A, the material scientist 102 interacts with two user interfaces provided by the computer-based system 104, namely a first user interface (UI) 103 and a second UI 105. In one computer operating system, the user interfaces 103, 105 are implemented as so-called windows. In a global network (Internet) server-based computer system, the user interfaces 103, 105 may be implemented as different tabs, screen views, or the like. The first UI 103 and the second UI 105 are presented on a display screen 107 of the computer-based system 104 for user interaction. Of course, the computer-based system 104 is not limited to providing two user interfaces (UIs).

In an exemplary embodiment, the first UI 103 includes a display of a wind turbine multi-scale model 109 of a physical (real-world) wind turbine. The second UI 105 includes a display of a 3D multi-scale model 106, i.e., a wind turbine blade multi-scale model 108 of a wind turbine blade of the wind turbine in the exemplary embodiment. It should be understood that the 3D system (or multi-scale model thereof) is not limited herein to the wind turbine blades or wind turbines described herein for illustrative purposes.

A wind turbine is a real-world object that converts wind energy into electricity using aerodynamic forces from the wind turbine blades, i.e., rotor blades that function like aircraft wings or helicopter rotor blades. Wind turbine blades are physical objects designed to be in use for 30 years or more. Physical tests are designed by structural test engineers to quantify the fatigue of such blades under cyclic loading. These physical tests can expose the wind turbine blades to mean stresses for wind conditions to determine the material failure limit after N cycles. In the exemplary embodiment of FIG. 1A, a computer-based system 104 is used to generate a multi-scale model 108 of the wind turbine blade that is an accurate representation of such a 3D system, allowing the material failure limit to be determined via computer simulation.

Wind turbine blades may be longer than aircraft wings. Thus, physical testing of such objects would use a significant amount of real estate to house the wind turbine blades and supplies to physically cycle the wind turbine blades to determine material failure limits. To avoid costly and time-consuming physical testing, a computer-based system 104 may be utilized to virtually determine such material failure limits.

A wind turbine blade is a composite structure involving several material types. For example, a wind turbine blade may have a composite skin that is an assembly of layers and core materials. Thus, each layer and core may have its own unique model in such a multi-scale model of a wind turbine blade. For example, the core may be a honeycomb structure as a non-limiting example. Resin from the curing process contributes to the overall stiffness of the core. A composite skin may combine several types of layers in variable ratios, including woven layers that may be in balance of 0 degrees to 90 degrees of tow (fiber). Models of such layers and core materials may be included in the set of representation models 118 disclosed above in connection with FIG. 1B.

In each unidirectional layer of the wind turbine blade, the fibers may be randomly distributed in the resin. Variabilities in the manufacturing process of the wind turbine blade, such as fiber diameter, fiber volume fraction, and minimum space between fibers, may result in artifacts, such as gases trapped during the resin curing process, creating a porous material. The artifact model 110 of FIG. 1B above may be generated at a given scale and may illustrate such artifacts as a non-limiting example. As disclosed above with reference to FIG. 1B, the artifact model 110 generated at a given scale illustrates the attributes 112, features 114, and artifacts 116 of the 3D system. According to an exemplary embodiment, a series of representation models 118 of a 3D system, such as a wind turbine blade or another 3D system, may be based on such artifact model 110. Each representation model in the series of representation models 118 may be constructed at a different scale. The different scales may include, as non-limiting examples, a chemical-material scale, a material-material scale, an engineering-design scale, an engineering-manufacturing-process scale, a system life scale, or a combination thereof.

Such modifications include mapping 120 the attributes 112, features 114, and artifacts 116 to display models in a set of display models 118 at higher or lower scales compared to a given scale. A non-limiting example of such a mapping 120 with two industry-driven sets of display models is disclosed as a non-limiting example for lithium-ion batteries and wind turbine blades. In a non-limiting example, there is both a need for attributes diffusion or tensile modulus, features voltage or strength, and manufacturing variability, such as cycles before a battery fails or time before a wind turbine blade has microcracks. A non-limiting example of such a mapping 120 is that for a lithium-ion battery, the model may map the diffusion rate of lithium ions from an electrolyte chemistry model. The diffusion rate may then also be mapped to a whole cell voltage/state of charge model for a lithium-ion battery to estimate the open cell voltage, and the mapping of artifacts may include mapping such models to include operating temperature or number of charge/discharge cycles to understand the changes in battery life. Another artifact mapping may be, by way of non-limiting example, mapping the change in open cell voltage with manufacturing characteristics (e.g., factory quality parameters such as, by way of non-limiting example, temperature, humidity, manufacturing time, location, etc.).

In the area of composite materials for wind turbine blades, the mapping 120 of attributes 112, features 114, and artifacts 116 to the representation models in the series 118 may include, as a non-limiting example, mapping the polymer structure used in the resin to a Young's modulus mechanical performance prediction combined with fiber spacing and orientation to failure stress for the formed part of the wind turbine blade. The artifacts 116 may be manufacturing characteristics, factory, batch of chemicals used, time and shift ID, and facility and process unit identifiers. In such an implementation, the aforementioned artifacts 116 may be, as a non-limiting example, mapped to the mean time between failures as the manufactured wind turbine blade. With further reference to FIGS. 1A and 1B, the mapping 120 allows bridging a given representation model of the series of representation models 118 at a given scale and a representation model at a higher or lower scale to generate an accurate 3D multi-scale model. The same simulation allows validation of the structure, i.e., as a non-limiting example, a wind turbine blade, or another type of 3D structure.

At the wind turbine blade scale, a global simulation can be performed by the computer-based system 104 according to the multi-scale simulation. The virtual model, i.e., the 3D multi-scale model 106, allows virtual modeling of the variability of the manufacturing process and allows a full virtual validation of the structure to be performed faster and at reduced cost compared to physical testing. According to an exemplary embodiment, the 3D multi-scale model 106 may be material class specific and subject to model and version control to ensure referential integrity.

As disclosed above with reference to FIGS. 1A and 1B, at least one processor of the computer-based system 104 automatically stores the artifact model 110 in association with the set of modified display models 118 in a database, thereby generating the 3D multi-scale model 106 of the 3D system. According to an example embodiment, the at least one processor may be further configured to automatically store the model information in the database.

The model information may represent, by way of non-limiting examples, starting information, training dataset information, learning method information, auxiliary data, measured or predicted data, or a combination thereof, to which the set of display models 118 corresponds. The model information may be associated in a database with the set of display models 118, the artifact models 110, or a combination thereof. Generating the artifact models 110 may include automatically identifying (determining) the attributes 112, features 114, and artifacts 116 via machine learning. The machine learning may include, by way of non-limiting examples, utilizing deep learning, adversarial learning, genetic or evolutionary methods, other modeling or segmentation classification approaches to modeling, or a combination thereof. The at least one processor may be further configured to perform machine learning on the systematic testing results of the set of samples.

The discrimination may be made at a given scale. However, it should be understood that the discrimination may incorporate other scale information, such as from lower scale details or higher scale performance characteristics. Lower scales are lower compared to the given scale and higher scales are higher compared to the given scale.

According to an exemplary embodiment, the at least one processor may be further configured to control the machine learning in a closed loop or according to at least one optimization criterion, such as Pareto optimization as determined via a Pareto optimization analysis as a non-limiting example. According to another exemplary embodiment, the at least one processor may be further configured to perform the machine learning iteratively based on a performance criterion, a convergence threshold, a quality metric, a limit or limits, or a combination thereof. The at least one processor may be further configured to determine whether the performance criterion, the convergence threshold, the quality metric, the limit or limits, or a combination thereof is satisfied via a closed loop or according to the at least one optimization criterion.

According to an exemplary embodiment, the at least one processor may be further configured to (i) generate at least one of the series of representation models 118 based on a manufacturing process, and (ii) generate the series of representation models 118 by utilizing features of a plurality of test coupons produced through the manufacturing process. The manufacturing process may be a characterized manufacturing process that is measured or characterized in some manner. In some cases, such as manufacturing scale-up or ramp-up, for example, process variables utilized by the manufacturing process may be systematically varied through designed experiments or tests, the results of which may be used as inputs to the development of a 3D multi-scale model 106 that is between the engineering scale and the manufacturing scale.

According to an example embodiment, each display model in the set of display models 118 may be constructed at a different scale of a plurality of scales, the plurality of scales including the given scale. The at least one processor may be further configured to generate a respective artifact model at each scale of the plurality of scales.

Although the exemplary embodiments disclosed herein may utilize test coupons that are standard for composite materials, it should be understood that such embodiments are not limited to test coupons and may, for example, utilize systematic test results of samples in place of test coupons.

According to an exemplary embodiment, the at least one processor may be further configured to train, during a training phase, the set of display models 118 based on at least one respective training data set (not shown). The at least one processor may be further configured to execute, during a run phase, the 3D multi-scale model 106 to generate predictions of onset of faults in the 3D system. The at least one processor may be further configured to improve the accuracy of predictions generated during the run phase by re-training, during a validation phase, the set of display models 118 and the artifact model 110 based on measurement data or prediction data input to the set of display models 118.

Exemplary embodiments disclosed herein include efficient methods that may use the machine learning disclosed above to generate chemically and physically realistic models to map atomic material-scale attributes to attributes measured on a set of test coupons made using a process from one or more materials. This is useful for improving simulations of material performance at bulk and nanometer scales and beyond. It is also useful for studying non-local and non-equilibrium phenomena and allowing understanding of long-scale attributes such as chemical decomposition in the environment. Although exemplary embodiments disclosed herein may be described with respect to composite systems such as wind turbine blades, it should be understood that the embodiments disclosed herein are not limited to composite systems and may be used in a broader context for any material, mixture, or formulated and manufactured system. Structures made of multiple materials may also be modeled using exemplary embodiments of the computer-implemented methods disclosed herein to better represent real-world materials.

2 is a flow diagram of an exemplary embodiment of a computer-implemented method (200) for generating a three-dimensional (3D) multi-scale model of a 3D system. The computer-implemented method includes generating (202) an artifact model indicative of attributes, features, and artifacts of the 3D system at a given scale. The computer-implemented method further includes modifying (204) a set of display models of the 3D system based on the generated artifact model. The modifying (206) includes mapping the attributes, features, and artifacts to a display model of the set of display models at a higher or lower scale compared to the given scale. The mapping bridges a given display model of the set of display models at the given scale and a display model at a higher or lower scale. The computer-implemented method further includes automatically storing (208) the artifact model in a database in association with the modified set of display models, thereby generating a 3D multi-scale model of the 3D system. The method then ends (210) in the exemplary embodiment.

The computer-implemented method 200 may further include automatically storing the model information in a database. The model information may represent starting information, training dataset information, learning method information, auxiliary data, measured or predicted data, or combinations thereof, to which the set of display models correspond. The model information may be associated in the database with the set of display models, artifact models, or combinations thereof. Generating the artifact models may include automatically identifying attributes, features, and artifacts via machine learning. The identification may be performed at a given scale. The machine learning may include utilizing deep learning, adversarial learning, genetic or evolutionary methods, other modeling or segmentation classification approaches to modeling, or combinations thereof.

Referring again to Figures 1A and 1B, according to an exemplary embodiment, quantum mechanical, atomistic, mesoscale, and constitutive models may be applied as inputs or features to the computer-based system 104 for use in generating a series of representation models 118. Mapping 120 allows for bridging of the nanoscale, engineering scale, and/or metrology scales.

According to an exemplary embodiment, the computer-based system 104 is self-updating, i.e., as new data is received, a set of display models 118 (also referred to herein as a set of iterative intermediate models) may be determined and then evaluated, ranked, and stored for future retrieval, e.g., by an adversarial machine learning process executed on the computer-based system 104.

According to another exemplary embodiment, test process and procedure information can be input to the computer-based system 104, which can be configured to use such information to categorize the model inputs into descriptors that can enhance the model of the system modeled by the computer-based system 104. According to an exemplary embodiment, the set of display models 118 can be stored in a database along with origins, data sets, methods, and results for authenticity and validation of tracking and tracing.

Connecting across scales may not be a new endeavor, but historically, all such attempts have been limited in scope, scalability, and accuracy. Key to the failure of existing approaches is the lack of good models that characterize artifacts introduced by testing and their incorporation into models that can be used to predict bottom-up performance from the material or top-down performance from the environment. The computer-based system 104 is a platform that enables the automatic generation of both attribute/performance models and artifact models. The computer-based system 104 enables the connection of models/simulations to predictions for design, development, and deployment.

The computer-based system 104 may use a machine learning approach to combine any or all of the following features, as non-limiting examples:
a) From materials (measured and virtual)
b) Material (formulations and recipes/implied process conditions)
c) Testing (results, operations, equipment)
d) Other scale models (reduced order/homogenized or full complexity)

According to an exemplary embodiment, the link between the method of deriving the set of display models 118 and their ranking (determined by validation on a test set) and the data used to construct them (the training set) may be maintained by the computer-based system 104, for example in a database, to allow for inter-test comparisons, updates of the display models in the set 118, and traceability and dependencies of decisions.

According to an exemplary embodiment, the computer-based system 104 enables the construction (updating existing or new) of the 3D multi-scale model 106 with both value predictions and uncertainties. The computer-based system 104 enables the model to be built in an automated manner, hierarchically, in a bottom-up approach, e.g., from atoms to molecules, to materials to distributions, to engineering elements, to products, and also allows the burden of top-down inputs (e.g., loads, cyclic effects, stresses, and environmental inputs) that drive fatigue effects and induce material changes (usually, but not always, failures) to be applied to the model of the system, such as a series of representation models 118 of the 3D multi-scale model 106 of the 3D system.

According to an exemplary embodiment, the computer-based system 104 may use multiple modeling approaches to iteratively match data (attributes) to features (other models, parameters, conditions, and loads), such as, but not limited to, generative adversarial networks (GANs), deep neural networks (DNNs), gene function approximation (GFA), and/or other multivariate statistical or classification approaches. Many such modeling approaches produce families of related models, and such models may be stored for reuse, for example, in a database. The computer-based system 104 may be configured to take such models and build approximate model variants of such models in a constructive form, applying reduced order modeling approaches to facilitate user interpretation of the impact of such variants.

According to an exemplary embodiment, the computer-based system 104 may be configured to use deep adversarial neural computing approaches together with iterative and probabilistic processes to obtain a 3D multiscale model 106, which may be a realistic computer model of a 3D system, bridging the scales of time, length, and complexity between atomistic, engineering, and manufacturing scale data.

According to another exemplary embodiment, the computer-implemented method may start with an untrained model network. Based on the data and information of a set of test coupons along with formulated material features from meso-, atomistic, nano-, and quantum-scale modeling, a set of display models may be developed and trained (step 1), such as the set of display models 118 disclosed above in connection with FIG. 1B. The models may then be optimized using adversarial approaches and methods to build improved models with both numerical accuracy and robust error estimates for different observed attributes and features (step 2). A deconvolution of the network may then be performed to identify the physical forms as equations of the underlying attribute relationships being developed (step 3). These new equations may then be re-optimized as features of the model network in step 1 (step 4). The results may then be stored in a data structure with meta tags for tracking during use (step 5). Different models, such as but not limited to parametric or data subset variants, may be generated by repeating the above process to obtain a final statistical average ensemble or composite model of the analyzed tests and processes. The model generation, training and optimization processes can be performed supervised or unsupervised, with integrated version control and master data generation.

Thus, according to an exemplary embodiment, a method for generating a 3D multi-scale model of a 3D system, such as a composite system, may include measuring a set of test coupons for a material system to be modeled, the test coupons produced from a given fabrication process, and physical properties of each of the test coupons. From the untrained model network, the method may generate one or more data models based on material features from one or more types of scale models from the given manufacturing process. The method may perform deep learning analysis on the set of physically measured attributes of the test coupons using an adversarial approach. In response to reaching a convergence threshold, the method may deconvolve the model using an inverse logic approach to extract a reduced-order model of the performance curve.

According to an exemplary embodiment, a notable improvement over existing methods for bridging models is that the exemplary embodiments disclosed herein separate the assumptions and approximations of artifacts introduced during model creation associated with mapping test samples and nanoscale data, such as test coupons, to the test coupon experimental level, as a non-limiting example, from the artifacts, assumptions and approximations of the manufactured product, its processes and sources of variability. Unique to this approach is that the individual scale gaps, from nano to meso, meso to macro, and macro to manufacturing, can be developed separately but connected in information space. Furthermore, mapping material, composition and system characteristics to product production performance and quality metrics based on a series of low-scale models allows for more accurate and robust model development.

Exemplary embodiments of the computer-based system 104 can use the extraction of model morphology to enhance, first, in a closed-loop approach where morphology is used in feature generation, and second, in higher-level subsequent models. The computer-based system 104 can be used to automatically generate robust 3D multi-scale models 106 with minimal user input. Such capabilities allow for rapid exploration of design parameter space and autonomous model development as new data accumulates from test or manufacturing systems, thereby providing an improved user experience due to the availability of multiple models. Exemplary embodiments disclosed herein can be used as a general-purpose tool in the modeling workflow and the 3DEXPERIENCE® platform, in both deterministic and feature analysis approaches. The 3DEXPERIENCE platform, available from Dassault Systems SE, is a collaborative environment platform that offers multiple software solutions.

According to an exemplary embodiment, a computer-implemented method may obtain a machine-learned and maintained model for the test coupon data and the manufactured product data. Chemical material model information and details of the fabrication process, state, and operation sequence may be encoded as features in the model, such as the multi-scale model 106 of FIG. 1A. The model may then autonomously perform deep learning analysis on the set of physically measured attributes using an adversarial approach. In response to a convergence threshold of the Pareto-optimal space being achieved, the model, its state, inputs, and outputs may be automatically stored in a data structure, and metadata may be generated to facilitate their extraction and unification. In a next step, the model form and functions may be deconvolved using an inverse learning approach with symbolic logic to extract the simplest reduced-order model of the performance curve. The computer-implemented method may then take the features or descriptors and perform the above model iterations on the manufactured part using the performance information extracted from the source system.

With reference to Figures 1A and 1B, an exemplary embodiment of a model, such as a display model of the series of models 118, can be constructed by the computer-based system 104 from nanoscale data having one type of material to test coupon measurement data using a modeling platform, such as the 3DEXPERIENCE platform, disclosed below in connection with Figure 3.

3 is a block diagram of an exemplary embodiment of a procedure 300 (a computer-implemented method) for building a display model. The display model may be referred to as an attribute model. In an exemplary embodiment, the procedure 300 is implemented, as a non-limiting example, using the 3DEXPERIENCE platform 332. The procedure 300 includes feature generation (334a), receiving an initial data set (334b), and utilizing an automated data pipeline (334c). The procedure 300 may explore simple architectures via an evolutionary approach and evaluate the accuracy, uncertainty, and speed (334e) of the display model (334d). The procedure 300 may train and evaluate multiple attributes via an adversarial approach (334f). The procedure 300 may screen multiple simple models (e.g., thousands, as a non-limiting example) with diverse architectures (334g). The procedure 300 may top the model stack (334h) and evaluate the same against benchmark predictions from previous model versions (334i). The procedure may select and retrain a top architecture (334j) and perform contextualization cleanup and augmentation to produce a representation model (334k). Elements of another procedure for building an attribute model are disclosed below in connection with FIG. 4.

Figure 4 is a block diagram of an exemplary embodiment of the analysis 444 elements that may be utilized by a complex procedure for constructing an attribute model from nanoscale data having multiple types of materials in different physical arrangements, such as layers, orientations, or distributions, for features 440, process fabrication characteristics 442, and test coupon measurement data. A simple procedure for constructing an attribute model from nanoscale data using one type of material is disclosed below in conjunction with Figure 5.

FIG. 5 is a flow diagram of an exemplary embodiment of a computer-implemented method 500 for building attribute models from nanoscale data using one type of material to machined part performance data. The method 500 may also be referred to as a method for bridging models. In an exemplary embodiment, such a model is a model associated with a hip ball joint utilized in hip replacements, as a non-limiting example.

The procedure 500 includes scanning (552) a real-world object using real-world scan data 503 as input to a model 506. The scanning may include recording spatial features of the real-world object, such as, by way of non-limiting example, a hip ball joint 556, as well as details regarding how the material of the hip ball joint 556 was created, features regarding adhesive fillers associated with the hip ball joint 556, microfeatures such as porosity, features regarding the coating(s) of the hip ball joint 556, by way of non-limiting example. It should be understood that the real-world object is a physical 3D object (system) and is not limited to the hip ball joint 556. Additionally, the real-world scan data 503 may include any information associated with the physical 3D object being modeled, such as, by way of non-limiting example, material characteristics and processes for creating the material. In an exemplary embodiment, the procedure 500 includes modeling 558 a 3D system, i.e., the hip ball joint 556 in the exemplary embodiment.

1A, 1B, 2, and 5, modeling 558 includes generating a 3D multi-scale model 506. The generation may be performed using a computer-based system 104, which may be configured to perform the computer-implemented method 200 of FIG. 2 disclosed above, as a non-limiting example. The attributes 112, features 114, and artifacts 116 of the artifact model 110 may include attributes, features, and artifacts contained in the real-world scan data 503. According to an exemplary embodiment, modeling 558 may include creating (562) a procedural model (e.g., a model of a process for making a material) and a cyclical model having parameters that control manufacturing variability of the process.

In the exemplary embodiment of FIG. 5, the procedure 500 may include a computer-implemented method of outputting 564 the 3D multi-scale model 506 to a simulator 566, i.e., simulating the use of the hip joint ball joint 556 using the model 506 to generate results 568 by subjecting the 3D multi-scale model 506 to manufacturing variability. Such results 568 may be utilized to perform a calibration 570 to calibrate the simulation model, i.e., the 3D multi-scale model 506 based on real-world test data. Such a calibration may improve the 3D multi-scale model 506 such that the 3D multi-scale model 506 represents a more realistic model of the 3D system, i.e., the hip joint ball joint 556 as a non-limiting example. The more realistic model may be used to determine, for example, material failure limits of the hip joint ball joint 556 as a non-limiting example. The procedure 500 may be implemented by an exemplary embodiment of a computer internal structure as disclosed below in connection with FIG. 6.

6 is a block diagram of an example of the internal structure of a computer 600 in which various embodiments of the present disclosure may be implemented. The computer 600 includes a system bus 602, which is a set of hardware lines used to transfer data between components of a computer or digital processing system. The system bus 602 is essentially a shared conduit connecting different elements of a computer system (e.g., processor, disk storage, memory, input/output ports, network ports, etc.), allowing information to be transferred between the elements. Coupled to the system bus 602 is an I/O device interface 604 for connecting various input and output devices (e.g., keyboard, mouse, display monitor, printer, speakers, microphone, etc.) to the computer 600. The network interface 606 allows the computer 600 to connect to various other devices attached to a network (e.g., a global computer network, a wide area network, a local area network, etc.). Memory 608 provides volatile or non-volatile storage for computer software instructions 610 and data 612 that may be used to implement embodiments of the present disclosure (e.g., methods 200, 300, 500), where volatile and non-volatile memory are examples of non-transitory media. Disk storage 613 also provides non-volatile storage for computer software instructions 610 and data 612 that may be used to implement embodiments of the present disclosure (e.g., methods 200, 300, 500). Central processor unit 618 is also coupled to system bus 602 and provides for the execution of computer instructions.

Further exemplary embodiments disclosed herein may be configured using a computer program product, e.g., a control may be programmed in software to implement the exemplary embodiments. Further exemplary embodiments may include a non-transitory computer-readable medium including instructions executable by a processor, which, when loaded and executed, cause the processor to complete the methods and techniques described herein. It should be understood that the elements of the block diagrams and flow diagrams may be implemented in software or hardware, such as through one or more arrangements of the circuitry of FIG. 6 disclosed above, or equivalents thereof, firmware, a combination thereof, or other similar implementations to be determined in the future.

Furthermore, the elements of the block diagrams and flow diagrams described herein may be combined or divided in any manner in software, hardware, or firmware. If implemented in software, the software may be written in any language capable of supporting the exemplary embodiments disclosed herein. The software may be stored in any form of computer-readable medium, such as random access memory (RAM), read-only memory (ROM), compact disc read-only memory (CD-ROM), etc. In operation, a general-purpose or application-specific processor or processing core loads and executes the software in a manner well understood in the art. It should be further understood that the block diagrams and flow diagrams may include more or fewer elements, may be arranged or oriented differently, or may be represented differently. It should be understood that implementations may define block diagrams, flow diagrams, and/or network diagrams illustrating the execution of the embodiments disclosed herein, as well as multiple block diagrams and flow diagrams.

Although exemplary embodiments have been specifically shown and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the embodiments encompassed by the appended claims.

Claims (20)

1. A computer-implemented method for generating a three-dimensional (3D) multi-scale model of a 3D system, the computer-implemented method comprising:
generating an artifact model illustrating attributes, features, and artifacts of a 3D system at a given scale , wherein generating the artifact model includes automatically identifying the attributes, features, and artifacts via machine learning, wherein the identifying is performed at the given scale;
modifying a set of display models of the 3D system based on the generated artifact model, the modifying including mapping the attributes, features and artifacts to a display model of the set of display models at a higher or lower scale compared to the given scale, the mapping bridging a given display model of the set of display models at the given scale and the display model at the higher or lower scale;
and automatically storing the artifact model in association with the modified set of display models in a database, thereby generating a 3D multi-scale model of the 3D system. further comprising automatically storing model information in the database;
the model information represents starting information, training dataset information, learning method information, auxiliary data, measured data, or predicted data, or a combination thereof, to which the set of display models correspond;
the model information is associated in the database with the set of representation models, the artifact models, or a combination thereof;
10. The computer-implemented method of claim 1. 2. The computer-implemented method of claim 1 , wherein the machine learning comprises utilizing deep learning, adversarial learning, genetic or evolutionary methods, other modeling or segmentation classification approaches to modeling, or combinations thereof. The computer-implemented method of claim 1 , further comprising performing the machine learning on a set of sample systematic test results. Controlling the machine learning in a closed loop or according to at least one optimization criterion;
performing the machine learning iteratively based on a performance criterion, a convergence threshold, a quality metric, a limit or limits, or a combination thereof;
2. The computer-implemented method of claim 1, further comprising: determining via the closed loop or in accordance with the at least one optimization criterion whether the performance criterion, convergence threshold, quality metric, limit or limits, or the combination thereof, is satisfied. The computer-implemented method of claim 1, further comprising: (i) generating at least one display model of the set of display models based on a manufacturing process; and (ii) generating the set of display models by utilizing features of a plurality of test coupons manufactured via the manufacturing process. The computer-implemented method of claim 1, wherein each display model of the set of display models is constructed at a different scale of a plurality of scales, the plurality of scales including the given scale, and the computer-implemented method further comprises generating a respective artifact model at each scale of the plurality of scales. training the set of representation models based on at least one respective training data set during a training phase;
executing the 3D multi-scale model during an execution phase, wherein said executing produces a prediction of an onset of a fault in the 3D system; and
2. The computer-implemented method of claim 1, further comprising: in a validation phase, improving accuracy of the predictions made in the execution phase by re-training the set of display models and artifacts based on measurement data or predicted data input for the set of display models. The computer-implemented method of claim 1, wherein the 3D system is a building system, component, material, or structure, and the structure includes i) a plurality of raw or intermediate materials, or ii) a mixture or formulation of the plurality of raw or intermediate materials. The computer-implemented method of claim 1, wherein the 3D system is a real-world system and each representation model in the series of representation models is constructed at a different scale, the different scales including a chemical-materials scale, a materials-materials scale, an engineering-design scale, an engineering-manufacturing-process scale, a system-life scale, or a combination thereof. 1. A computer-based system for generating a three-dimensional (3D) multi-scale model of a 3D system, the computer-based system comprising:
At least one memory;
at least one processor coupled to the at least one memory, the at least one processor comprising:
generating an artifact model illustrating attributes, features, and artifacts of a 3D system at a given scale , wherein generating the artifact model includes automatically identifying the attributes, features, and artifacts via machine learning, wherein the identifying is performed at the given scale;
modifying a set of display models of the 3D system based on the generated artifact model, the modifying including mapping the attributes, features and artifacts to a display model of the set of display models at a higher or lower scale compared to the given scale, the mapping bridging a given display model of the set of display models at the given scale and the display model at the higher or lower scale;
and automatically storing the artifact model in a database in association with the modified set of display models, thereby generating a 3D multi-scale model of the 3D system. the at least one processor is further configured to automatically store model information in the database;
the model information represents starting information, training dataset information, learning method information, auxiliary data, measured data, or predicted data, or a combination thereof, to which the set of display models correspond;
the model information is associated in the database with the set of representation models, the artifact models, or a combination thereof;
12. The computer-based system of claim 11. The computer-based system of claim 11 , wherein the machine learning includes utilizing deep learning, adversarial learning, genetic or evolutionary methods, other modeling or segmentation classification approaches to modeling, or combinations thereof. The computer-based system of claim 11 , wherein the at least one processor is further configured to perform the machine learning on systematic test results of a set of samples. the at least one processor:
Controlling the machine learning in a closed loop or according to at least one optimization criterion;
performing the machine learning iteratively based on a performance criterion, a convergence threshold, a quality metric, a limit or limits, or a combination thereof;
The computer-based system of claim 11, further configured to: determine whether the performance criterion, convergence threshold, quality metric, limit or limits, or any combination thereof, is satisfied via the closed loop or in accordance with the at least one optimization criterion. the at least one processor:
12. The computer-based system of claim 11, further configured to: (i) generate at least one display model of the series of display models based on a manufacturing process; and (ii) generate the series of display models by utilizing features of a plurality of test coupons produced via the manufacturing process. The computer-based system of claim 11, wherein each display model of the set of display models is constructed at a different scale of a plurality of scales, the plurality of scales including the given scale, and the at least one processor is further configured to generate a respective artifact model at each scale of the plurality of scales. the at least one processor:
training the set of representation models based on at least one respective training data set during a training phase;
and executing the 3D multi-scale model to generate a prediction of an onset of a fault in the 3D system.
12. The computer-based system of claim 11, further configured to: during a validation phase, improve accuracy of the predictions made during the execution phase by re-training the set of display models and artifact models based on measurement data or prediction data input for the set of display models. The computer-based system of claim 11, wherein the 3D system is a building system, component, material, or structure, and the structure includes i) a plurality of raw or intermediate materials, or ii) a mixture or formulation of the plurality of raw or intermediate materials. 1. A non-transitory computer-readable medium for generating a three-dimensional (3D) multi-scale model of a 3D system, the non-transitory computer-readable medium, when loaded and executed by at least one processor, causing the at least one processor to:
generating an artifact model illustrating attributes, features, and artifacts of a 3D system at a given scale , wherein generating the artifact model includes automatically identifying the attributes, features, and artifacts via machine learning, wherein the identifying is performed at the given scale;
modifying a set of display models of the 3D system based on the generated artifact model, the set of modifications including mapping the attributes, features, and artifacts to a display model of the set of display models at a higher or lower scale compared to the given scale, the mapping bridging a given display model of the set of display models at the given scale and the display model at the higher or lower scale;
and automatically storing the artifact model in a database in association with the modified set of display models, thereby generating a 3D multi-scale model of the 3D system.

Images